Federated Learning is a promising branch of Machine Learning which tries to create model without using client’s data unlike traditional Machine Learning algorithms. This approach helps to overcome the challenge of data privacy. Local models are aggregated in the server after trained on edge devices. After edge devices send their trained model, these models are aggregated on the server by aggregation algorithms. One of such algorithms is Federated Averaging. Problem with this algorithm is that it is naive approach which only takes number of training samples for each client as weight. In this work, we modified this algorithm by adding Genetic Algorithm. The weights of the clients are to be chosen by Genetic Algorithm which tries to improve its performance in each generation. The new approach was tested on two different datasets. Obtained results showed that suggested approach increases the performance of global model by 1–2%. Though the improvement was achieved, observations show that there are still room for further improvement.